/*
* $RCSfile: VisualNeuronNetwork.java,v $
* $Revision: 1.2 $
* $Date: 2007/08/18 09:49:56 $
* $Author: wojna $
*
* Copyright (C) 2002 - 2007 Logic Group, Institute of Mathematics, Warsaw University
*
* This file is part of Rseslib.
*
* Rseslib is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 3 of the License, or
* (at your option) any later version.
*
* Rseslib is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
package rseslib.processing.classification.neural;
import javax.swing.JPanel;
import javax.swing.BoxLayout;
import java.awt.event.ActionEvent;
import java.awt.event.ActionListener;
import java.util.Properties;
import javax.swing.*;
import rseslib.processing.classification.VisualClassifier;
import rseslib.structure.attribute.Header;
import rseslib.structure.data.DoubleData;
import rseslib.structure.table.DoubleDataTable;
import rseslib.system.PropertyConfigurationException;
import rseslib.system.Report;
import rseslib.system.progress.Progress;
/**
* Wizualizacja implementacji sieci neuronowej w bibliotece rseslib. Wersja dwuwatkowa
* zostala zakomentowana. Jej odzyskanie polega na odkomentowaniu:
* w definicjach: private Task task
* w metodach: zmiana w startTrain()
* w kalsach" odkomentwanie klasy Task
* @author Damian Wojcik
*
*/
public class VisualNeuronNetwork extends NeuronNetwork implements VisualClassifier,
ActionListener {
public VNNPanel nnView, nnViewOne;
public AddNodesDialog addNodesView;
private Task task;
private JPanel canvas = null;
private JPanel canvasOne = null;
/**
* Konstruktor
*
* @param prop wlasciwosci klasyfikatora
* @param trainTable tablelka treningowa
* @param prog postep
* @throws PropertyConfigurationException
*/
public VisualNeuronNetwork(Properties prop, DoubleDataTable trainTable, Progress prog)
throws PropertyConfigurationException, InterruptedException
{
super(prop, trainTable, prog);
if (getBoolProperty("showTraining")) startTrain(); /* wersja dwuwatkowa*/
}
public void startTrain() {
task = new Task();
task.execute();
}
/*
* (non-Javadoc)
*
* @see java.awt.event.ActionListener#actionPerformed(java.awt.event.ActionEvent)
*/
public void actionPerformed(ActionEvent event) {
if (nnView != null) {
if (event.getSource() == nnView.TrainButton) {
/*rozpoczecie trenowania sieci*/
nnView.showClassification = false;
startTrain();
return;
}
if (event.getSource() == nnView.AddNodes) {
/*wyswietlenie panelu dodania perceptronow*/
if (addNodesView == null) {
addNodesView = new AddNodesDialog(this);
}
addNodesView.setVisible(true);
}
}
}
private void init(JPanel canv) {
nnView = new VNNPanel(networkStructure, bestEngine.layers, trainTable, true);
nnView.TrainButton.addActionListener(this);
nnView.AddNodes.addActionListener(this);
JScrollPane scrollPane = new JScrollPane(nnView);
canv.add(scrollPane);
canv.setLayout(new BoxLayout( canv, BoxLayout.Y_AXIS));
}
private void initOne(JPanel canv) {
nnViewOne = new VNNPanel(networkStructure, bestEngine.layers, trainTable, false);
JScrollPane scrollPane = new JScrollPane(nnViewOne);
canv.add(scrollPane);
canv.setLayout(new BoxLayout( canv, BoxLayout.Y_AXIS));
}
public void draw(JPanel canv) {
if (! canv.equals(canvas)) {
canvas = canv;
init(canvas);
nnView.setNotShowClassification();
canvas.revalidate();
canvas.setVisible(true);
canvas.repaint();
}
}
public double drawClassify(JPanel canv, DoubleData obj) {
double ret = classify(obj);
if (! canv.equals(canvasOne)) {
canvasOne = canv;
initOne(canvasOne);
nnViewOne.setShowClassification();
canvasOne.revalidate();
canvasOne.setVisible(true);
canvasOne.repaint();
return ret;
}
nnViewOne.setShowClassification();
canvasOne.repaint();
return ret;
}
public Header attributes() {
// TODO Auto-generated method stub
return null;
}
/**
*
* @param row - numer ukrytej warstwy
* @param no - liczba neuronow do dodania
*/
public void addNodes(int row, int no) {
if ((row > 0) && (row < bestEngine.layers.size())) {
for (int i=0; i<no; i++) {
//bestEngine.changeNetwork(row-1);
networkStructure[row]++;
}
bestEngine = new NeuronNetworkEngine(trainTable, trainData, validateData, networkStructure.length - 2, networkStructure);
if (nnView != null) nnView.updateModel(networkStructure, bestEngine.layers);
if (nnViewOne != null) nnViewOne.updateModel(networkStructure, bestEngine.layers);
}
}
/**
* Uczenie sieci
*
* @param trainTable -
* tabelka z danymi treningowymi
*/
protected void learn() throws PropertyConfigurationException {
// wyniki poszczegolnych silnikow na danych walidacyjnych
double result = 0;
// najnajlepszy do tej pory uzyskany wynik
double max_absolute_result = -1;
double max_result = -1;
// zserializowany najlepszy silnik
Object best_perceptrons_weights = null;
// numer rundy w ktorej osiagnelismy najlepszy wynik
int best_round = 0;
for (int i = 1; i < Global.MAX_REPEAT_COUNT; i++) {
Report.debugnl("Tura " + i);
if (nnView!=null)
{
nnView.TrainButton.setEnabled(false);
nnView.AddNodes.setEnabled(false);
}
// czas rozpoczecia rundy
lastRound = System.currentTimeMillis();
// czy jeszcze jakies silniki powinny byc uczone
boolean shouldLearnMore = false;
bestEngine.learn();
// wynik na danych walidacyjnych
result = bestEngine.targetRatio();
Report.debugnl("Wynik enginu : " + result);
// czy jestesmy najnajlepsi
if (result > max_absolute_result) {
max_absolute_result = result;
best_perceptrons_weights = bestEngine.storeData();
}
if (result > max_result) {
best_round = i;
max_result = result;
}
// czy wymagamy uczenia
if (result < Global.DEST_TARGET_RATIO)
shouldLearnMore = true;
// raport postepu
reportStep(System.currentTimeMillis() - startTime);
Report.debugnl("Czas dzialania "
+ (System.currentTimeMillis() - startTime) / 1000
+ " sek.");
// usuwanie silnika, ktorego dzialanie nie przynosi juz efektow,
// startowanie w zamian nowego
if (i - best_round > Global.GRACE_LEARN_PERIOD) {
Report.debugnl("Usunieto bezuzyteczny silnik");
bestEngine = new NeuronNetworkEngine(trainTable, trainData, validateData, networkStructure.length - 2, networkStructure);
if (nnView != null) {
nnView.updateModel(networkStructure, bestEngine.layers);
}
if (nnViewOne != null) {
nnViewOne.updateModel(networkStructure, bestEngine.layers);
}
best_round = i;
max_result = -1;
}
else
{
if (nnView != null) nnView.setNotValidShapes();
if (nnViewOne != null) nnViewOne.setNotValidShapes();
}
// czas trwania ostatniej rundy
long lastRoundTime = System.currentTimeMillis() - lastRound;
// jesli konczy sie czas lub nie ma sie czego uczyc to konczymy
long timeElapsed = System.currentTimeMillis() - startTime;
if (timeElapsed + 1.4 * lastRoundTime > this.timeLimit
|| !shouldLearnMore || reportStep(timeElapsed))
break;
}
// przywrocenie najlepszych wag
bestEngine.restoreData(best_perceptrons_weights);
if (nnView != null)
{
nnView.TrainButton.setEnabled(true);
nnView.AddNodes.setEnabled(true);
nnView.setNotValidShapes();
}
if (nnViewOne != null) nnViewOne.setNotValidShapes();
Report.debugnl("Walidacja najlepszego daje wynik "
+ bestEngine.targetRatio());
reportStep(this.timeLimit);
repaint();
}
public void repaint() {
if (canvas != null) {
canvas.repaint();
}
if (canvasOne != null) {
canvasOne.repaint();
}
}
/* watek dla klasyfikacji */
class Task extends SwingWorker<Void, Void> {
public Void doInBackground() {
startTime = System.currentTimeMillis();
try {
learn();
} catch (PropertyConfigurationException e) { }
return null;
}
public void done() {
;
}
}
}
